It is important to clearly know your limitations. Executive managers should be well aware of the current state of the art as even small changes in project task can cost months if not quarters of additional development.

To avoid this fallacies we maximize transparency:

• Highlight the risky parts of developing component
  
  
• Clearly state metrics or formulate limitations of the project
  
  
• Use clear checkpoints of development
  
  
• Perform our own management of the team to prevent miscommunication

Our team of skilled professionals and experts is advantageous for AI component development. We are continually training in developing new ML solutions without spending time on usual tasks the IT division has to work on.

We encapsulated our knowledge and experience in a set of off the shelf solutions that can be configured to solve particular business tasks. Our components are completely open-ended, easily customizable and can be integrated into any infrastructure.

Adding human operators to ML system can be a game changer in various cases. Assessors can be sent for data labeling, model relearning and checkup before sending.

CI/CD is not enough for model zoo handling, to work with ML and data models in production one needs to employ automated model quality testing, anomaly detection and so on

What we use in our systems and projects:

• Model serving
  
• Relearning and classification cycle
• Automated quality testing